The preceding summary of realization rules focusses on properties that are shared by all or most realizational models; the elaborations proposed in various extended models are discussed in Section 6.4. The initial description of basic rule types also introduces common organizational aspects of realizational models. In all models, feature bundles, representing the properties of paradigm cells or syntactic preterminals, specify a set of distinctive features. Realization rules spell out these features by successively modifying an initial root or stem form. Exponence rules express direct mappings between features and forms. Referral rules, in those models that incorporate them, function in effect as ‘second order’ exponence rules, relating multiple spell-outs.
Although realizational models are not normally regarded as derivational in the same sense as transformational accounts, realizational rules apply successively to spell out a bundle. The first rule in the sequence directly applies to a lexical root or stem, and subsequent rules apply in a fixed order, with each rule in the sequence defining the input to the following rule. At each point in the analysis of a complex form, there is usually a number of inflectional ‘choices’. These individual choices are expressed by ‘competing’ exponence rules.
The principles that regulate rule ordering and rule competition define the syntagmatic and paradigmatic dimensions of a realizational model. Rule ordering is imposed by organizing rules into sets, which, following Anderson (1992), are usually termed blocks. An extrinsic order is defined on blocks, so that rules in an ‘earlier’ block apply before those in a ‘later’ block. There is some freedom in how blocks are ordered. In Anderson’s analyses of Georgian and Potawotami, prefixal blocks are ordered before suffixal blocks, which appears to reflect the linear structure of formatives, proceeding from left-to-right. In contrast, the stem structure assigned to elelykete by Matthews (1991:178) exhibits a nested organization, corresponding to a block structure in which earlier blocks occur closest to the root. This order can again be reversed so that earlier blocks are more peripheral. The choice between these alternatives raises a number of further questions, which are taken up in Section 6.3.1.
Within each block, the competition between rules is usually resolved by principles that assign priority to the most specific applicable rule. Informally, rules are applicable to a feature bundle that contains all of the features that they spell out, and one rule is more specific than a second rule when it properly contains the features of the second. A notion of rule applicability defined in terms of feature inclusion rather than simple compatibility avoids the need to constrain realization rules through the use of negative or contrasting feature values (of the kind that prevent the ‘overapplication’ of IP processes).
The role that specificity-based competition plays in constraining the application of general rules embodies a distinctive realizational perspective on the nature of morphological variation. Each realization rule applies to a natural class of feature bundles, where the generality of the class is inversely proportional to the specificity of the rule. Patterns of variation associated with an apparently non-natural class of bundles are generally assumed to reflect the interaction of a general rule with more specific rules. The general rule is applicable in principle to a natural class of bundles. However it is only realized in a subset of these bundles because it is preempted by more specific rules in others.
In sum, realizational models adopt the hypothesis that patterns that characterize non-natural classes reflect the interaction of natural class patterns of varying levels of specificity. Extending this approach to different types of patterns with a nonnatural distribution creates challenges for a realizational model. Some familiar challenges are summarized in Sections 6.3.1-6.3.4.
-  However, as Olivier Bonami has drawn to my attention (p.c.), Anderson provides no explicitjustification for the relative ordering of prefixal and suffixal agreement blocks, and elsewhere orders thesuffixal theme vowel block before the prefixal agreement block. So the relation between linear order andblock ordering remains somewhat less clear in Anderson (1986,1992) than in the model of Crysmannand Bonami (2012).